Intelligent RACH Access strategies for M2M Traffic over Cellular Networks

This thesis investigates the coexistence of Machine-to-Machine (M2M) and Human-to-Human (H2H) based traffic sharing the Random Access Channel (RACH) of an existing cellular network and introduced a Q-learning as a mean of supporting the M2M traffic. The learning enables an intelligent slot selection strategy in order to avoid collisions amongst the M2M users during the RACH contest. It is also applied so that no central entity is involved in the slot selection process, to avoid tampering with the existing network standards. The thesis also introduces a novel back-off scheme for RACH access which provides separate frames for M2M and conventional cellular (H2H) retransmissions and is capable of dynamically adapting the frame size in order to maximise channel throughput. A Frame ALOHA for a Q-learning RACH access scheme is developed to realise collision- free RACH access between the H2H and M2M user groups. The scheme introduces a separate frame for H2H and M2M to use in both the first attempt and retransmissions. In addition analytical models are developed to examine the interaction of H2H and M2M traffic on the RACH channel, and to evaluate the throughput performance of both slotted ALOHA and Q-learning based access schemes. In general it is shown that Q-learning can be effectively applied for M2M traffic, significantly increasing the throughput capability of the channel with respect to conventional slotted ALOHA access. Dynamic adaptation of the back-off frames is shown to offer further improvements relative to a fixed frame scheme. Also the FA-QL-RACH scheme offers better performance than the QL-RACH and FB-QL-RACH scheme.